i965/fs: Lower 32x32 bit multiplication on BXT.

[mesa.git] / src / mesa / drivers / dri / i965 / brw_nir.c

diff --git a/src/mesa/drivers/dri/i965/brw_nir.c b/src/mesa/drivers/dri/i965/brw_nir.c
index 8700cb71ad409b1e5a812c138ff80ece69ef840a..b5788fa2e33b0ee0192474cff2454f8d1df7330b 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_nir.c
+++ b/src/mesa/drivers/dri/i965/brw_nir.c
@@ -27,19 +27,27 @@
 #include "program/prog_to_nir.h"
 
 static void
-nir_optimize(nir_shader *nir)
+nir_optimize(nir_shader *nir, bool is_scalar)
 {
    bool progress;
    do {
       progress = false;
       nir_lower_vars_to_ssa(nir);
       nir_validate_shader(nir);
-      nir_lower_alu_to_scalar(nir);
-      nir_validate_shader(nir);
+
+      if (is_scalar) {
+         nir_lower_alu_to_scalar(nir);
+         nir_validate_shader(nir);
+      }
+
       progress |= nir_copy_prop(nir);
       nir_validate_shader(nir);
-      nir_lower_phis_to_scalar(nir);
-      nir_validate_shader(nir);
+
+      if (is_scalar) {
+         nir_lower_phis_to_scalar(nir);
+         nir_validate_shader(nir);
+      }
+
       progress |= nir_copy_prop(nir);
       nir_validate_shader(nir);
       progress |= nir_opt_dce(nir);
@@ -92,21 +100,30 @@ brw_create_nir(struct brw_context *brw,
    nir_split_var_copies(nir);
    nir_validate_shader(nir);
 
-   nir_optimize(nir);
+   nir_optimize(nir, is_scalar);
 
    /* Lower a bunch of stuff */
    nir_lower_var_copies(nir);
    nir_validate_shader(nir);
 
    /* Get rid of split copies */
-   nir_optimize(nir);
+   nir_optimize(nir, is_scalar);
+
+   if (is_scalar) {
+      nir_assign_var_locations_direct_first(nir, &nir->uniforms,
+                                            &nir->num_direct_uniforms,
+                                            &nir->num_uniforms,
+                                            is_scalar);
+      nir_assign_var_locations(&nir->outputs, &nir->num_outputs, is_scalar);
+   } else {
+      nir_assign_var_locations(&nir->uniforms,
+                               &nir->num_uniforms,
+                               is_scalar);
 
-   nir_assign_var_locations_direct_first(nir, &nir->uniforms,
-                                         &nir->num_direct_uniforms,
-                                         &nir->num_uniforms,
-                                         is_scalar);
+      foreach_list_typed(nir_variable, var, node, &nir->outputs)
+         var->data.driver_location = var->data.location;
+   }
    nir_assign_var_locations(&nir->inputs, &nir->num_inputs, is_scalar);
-   nir_assign_var_locations(&nir->outputs, &nir->num_outputs, is_scalar);
 
    nir_lower_io(nir, is_scalar);
 
@@ -126,7 +143,7 @@ brw_create_nir(struct brw_context *brw,
    nir_lower_atomics(nir);
    nir_validate_shader(nir);
 
-   nir_optimize(nir);
+   nir_optimize(nir, is_scalar);
 
    if (brw->gen >= 6) {
       /* Try and fuse multiply-adds */
@@ -159,9 +176,14 @@ brw_create_nir(struct brw_context *brw,
       nir_print_shader(nir, stderr);
    }
 
-   nir_convert_from_ssa(nir, true);
+   nir_convert_from_ssa(nir, is_scalar);
    nir_validate_shader(nir);
 
+   if (!is_scalar) {
+      nir_lower_vec_to_movs(nir);
+      nir_validate_shader(nir);
+   }
+
    /* This is the last pass we run before we start emitting stuff.  It
     * determines when we need to insert boolean resolves on Gen <= 5.  We
     * run it last because it stashes data in instr->pass_flags and we don't
@@ -180,3 +202,42 @@ brw_create_nir(struct brw_context *brw,
 
    return nir;
 }
+
+enum brw_reg_type
+brw_type_for_nir_type(nir_alu_type type)
+{
+   switch (type) {
+   case nir_type_unsigned:
+      return BRW_REGISTER_TYPE_UD;
+   case nir_type_bool:
+   case nir_type_int:
+      return BRW_REGISTER_TYPE_D;
+   case nir_type_float:
+      return BRW_REGISTER_TYPE_F;
+   default:
+      unreachable("unknown type");
+   }
+
+   return BRW_REGISTER_TYPE_F;
+}
+
+/* Returns the glsl_base_type corresponding to a nir_alu_type.
+ * This is used by both brw_vec4_nir and brw_fs_nir.
+ */
+enum glsl_base_type
+brw_glsl_base_type_for_nir_type(nir_alu_type type)
+{
+   switch (type) {
+   case nir_type_float:
+      return GLSL_TYPE_FLOAT;
+
+   case nir_type_int:
+      return GLSL_TYPE_INT;
+
+   case nir_type_unsigned:
+      return GLSL_TYPE_UINT;
+
+   default:
+      unreachable("bad type");
+   }
+}